FIELD NOTE · 2026-05-18 · POST-LAUNCH EVIDENCE · ACY SECURITIES RG 268

Post-launch iteration — what the ACY cross-border disclosure surface looked like eighteen months after v1

A senior PD case study should not end at “we shipped it.” The interesting part is what users did with it that I did not predict, what I iterated, and the metric I expected to move that never did. Eighteen months on one regulatory surface, three design versions, one honest negative result.

11 min Read time
3 versions Of the RG 268 disclosure surface
18 months From v1 ship to v3 stable
1 metric That didn’t move (named in Section 4)
TL;DR — 60-second summary

Shipping the v1 of the ASIC RG 268 cross-border disclosure surface was the easiest part. The next eighteen months — two iteration cycles, one user workaround pattern I had not predicted, and one metric I expected to move that never did — were where the actual design work happened.

ACY Securities is an ASIC-regulated broker (AFSL 403863) serving retail and institutional traders across forty-plus jurisdictions. ASIC RG 268 requires meaningful engagement with jurisdiction-specific risk and cost disclosures before each retail transaction, not once at onboarding. The v1 surface I shipped in mid-2023 was a modal interstitial pattern that scored well on the compliance review but failed the durability test. Months three and four of post-launch behavioural data showed users had developed a muscle-memory dismissal pattern; the disclosure was being seen but not read. The v2 redesign moved to a persistent header strip, which failed for a different reason. The v3, shipped at month thirteen and stable through month eighteen, is a contextual inline pattern computed at order time.

This page documents all three versions, the user behaviour that drove the iterations, and the secondary metric that I expected to improve and never did. Microsoft’s brutal-review composite specifically flagged that my portfolio “ends too early — show me six months after launch.” The answer is here.

Section 1 — The v1 that shipped (months 0–3): modal interstitial

What it was. A full-screen modal interstitial that appeared before the order-confirmation step on any retail transaction crossing jurisdictional boundaries. The modal carried the RG 268-required disclosure text computed from the client’s declared jurisdiction, the product issuer’s jurisdiction, and the relevant AFSL framework. Three buttons at the bottom: Acknowledge & Proceed, Review Full Disclosure, Cancel.

Why we shipped it as v1. The modal pattern was the lowest-risk option from a regulatory perspective. ASIC’s engagement standard is met by a friction surface that cannot be bypassed; a modal satisfies that contractually. Legal signed off in one review cycle. Compliance liked the audit-trail discipline — every dismissal is a logged event with the disclosed text version, the user’s declared jurisdiction, and a server timestamp.

What the v1 numbers looked like for the first ninety days. Order abandonment lifted from the pre-RG-268 baseline of approximately 1.4% to 3.2% — an expected friction cost. The audit trail was clean. Compliance closed the project as complete. From the launch retrospective view, this was a win: we had shipped on time, satisfied the regulator, and the cost was a manageable two-point lift in abandonment that the business absorbed as the price of compliance.

That was the story for ninety days. Then the behaviour shifted in a way the launch metrics did not capture.

Section 2 — What changed at month three (the workaround I had not predicted)

The cohort-level data started telling a different story. By month three, the time-to-acknowledge on the modal had collapsed for repeat users. Median dismissal time on the modal’s first appearance: 18 seconds (consistent with reading the disclosure). Median dismissal time on the twentieth appearance for a returning trader: 0.6 seconds — faster than the time it would take to read the headline of the disclosure, let alone the body text. Power users had developed muscle memory: the modal appears, the right hand is already on the keyboard, Enter dismisses Acknowledge & Proceed before the eye has even reached the modal’s title.

The compliance review missed this because the audit trail looked the same. Every dismissal still logged with timestamp, disclosure version, and user jurisdiction. From the audit-trail perspective, the user had “engaged” with the disclosure. From the regulatory-intent perspective — RG 268’s requirement is meaningful engagement — the user had not. The regulator’s likely position, if this had come up in an actual ASIC review, would have been that a 0.6-second median dismissal time on a 400-word disclosure does not satisfy the engagement test.

I had assumed the friction was the engagement. The pre-launch design hypothesis was that the modal’s un-bypassable nature would force engagement. What actually happened: the modal’s un-bypassable nature trained users to dismiss faster. The friction was real for the first ten appearances and decayed asymptotically after that. By appearance forty, the modal was wallpaper.

What I did not predict. I had not modelled the temporal decay of muscle-memory dismissal. The pre-launch usability study (n = 12 retail traders, two sessions each) captured the first-appearance read time correctly but did not run sessions long enough to capture appearances five, ten, fifty. The modal as a pattern works once. The modal as a pattern at the twentieth appearance is a different product and I had not designed for it.

Section 3 — Three iterations over eighteen months

Version 2: persistent header strip (months 4–9). The hypothesis: if the modal was being trained-into-wallpaper because of its modal-ness, replace it with a passive surface that does not interrupt and does not train dismissal behaviour. The v2 was a persistent header strip across the top of every trading screen showing a one-line distilled version of the RG 268 disclosure computed for the user’s active context. No dismissal button — the strip is always there, like the regulator’s name on a brokerage’s footer.

The v2 numbers told us a different failure mode. Order abandonment dropped back to near-baseline (1.5%) because there was no longer a friction surface to abandon at. But the engagement metric we instrumented this time around — eye-tracking on a recruited n = 18 sub-cohort — showed that the strip received a fixation in the user’s first session (an average of 1.1 seconds of dwell time) and then almost zero attention from session three onward. The strip was visible but not seen. Eye-tracking confirmed users had banded the strip into the chrome zone they ignore by default, the same way they ignore the browser toolbar.

Compliance was unhappy with the v2 for a different reason than they would have been unhappy with the v1: the persistent strip’s text was generic by necessity (the strip is on every screen, so the disclosure has to be the union of all jurisdictions). Generic disclosure is weaker than jurisdiction-specific disclosure for the RG 268 engagement test. The v2 was less interruptive but its content was less regulatorily-precise.

Version 3: contextual inline at order time (months 10–18 and stable since). The insight from v1 and v2 combined: the disclosure should appear at the moment of regulatory relevance — when the user is committing to a specific cross-border transaction — and should be specific to that transaction, not generic. The v3 is a single line of computed disclosure that appears in the order ticket itself, between the size input and the submit button, computed from the client’s declared jurisdiction times the product issuer times the relevant AFSL framework. The user has to look at it because their eye is already there to confirm the order size. There is no separate dismissal event; the act of pressing Submit is the acknowledgment.

The v3 numbers stabilised between months thirteen and eighteen at: 1.6 seconds average sustained read time on the disclosure line (a 3x improvement over the v2 strip), 0.4% incremental order abandonment over the pre-RG-268 baseline (a 7x improvement over the v1 modal), full text specificity per transaction (passing the RG 268 engagement test in a way the v2 generic strip did not), and zero additional clicks compared to the pre-RG-268 order flow.

The v3 was the version that absorbed two subsequent RG 268 revisions (an ASIC guidance update in 2024 and a cross-border product re-classification in 2025) without a structural redesign — only token-level changes to the disclosure-text computation rules. This is the regulatory-rewrite-absorption pattern referenced elsewhere in the portfolio; the v3 design is the canonical example of how token-first absorption actually works in practice.

Section 4 — The metric that didn’t move (honest negative result)

I expected average AUM per new client to lift after the v3 stabilised. The reasoning: if cross-border disclosure friction was being managed without abandoning orders or losing engagement quality, new clients in the post-v3 window should be less spooked by jurisdiction prompts during onboarding, complete their first multi-jurisdiction allocation faster, and end up with broader portfolio diversification reflected in higher per-client AUM. The pre-launch business case for the v3 had a soft projection of 5–10% AUM lift per new client over an 18-month window post-stabilisation.

That lift did not materialise. Average AUM per new client in the post-v3-stable cohort (months 13–18) was statistically indistinguishable from the pre-v1 baseline. The 5–10% projection was wrong. The post-v3 data shows the AUM number is driven by macroeconomic factors (FX volatility, equity market conditions, broader retail capital flow into the broker category) and is largely insensitive to the disclosure-surface UX. The disclosure UX was a real product win on the regulatory-engagement and abandonment-cost axes — just not on the AUM axis.

Why the honest negative result matters. The pre-launch business case had attributed AUM lift to the UX improvement. If I had stayed quiet about the AUM number not moving, the portfolio headline could have read “v3 lifted AUM per client by X%.” The portfolio would have been stronger as a sales artifact and weaker as a senior-PD evidence document. The choice to report the negative result is the discipline that separates a designer who is selling their work from a designer who is documenting it.

What the post-v3 data did support. The order-abandonment cost dropped from 3.2% (v1 steady-state) to 0.4% (v3 steady-state), which over 100K+ traders at $2B+ daily volume is a meaningful dollar number even at thin retail-broker margins. The compliance-engagement quality (sustained read time + jurisdiction-specific text + audit-trail completeness) passed an internal mock-audit cleanly. Those two gains are real and externally-attestable. The AUM gain did not exist.

Section 5 — What I’d track from day 1 if running this today

If I were specifying the same RG 268 disclosure project today, the instrumentation plan would change in six directions:

1. Pre-register the post-launch metrics. Before v1 ships, commit in writing to which metrics will be reported at month 3, month 6, month 12. Distinguish between regulatory-engagement metrics (read time, dismissal time, audit trail completeness) and business-outcome metrics (AUM, order abandonment, per-client revenue). The pre-registration prevents post-hoc cherry-picking when the actual results come in.

2. Instrument temporal decay from day 1. The single biggest missing dimension in the v1 analytics was the appearance-count-versus-dismissal-time curve. If I had had that chart at month two, the muscle-memory problem would have been visible eight weeks earlier. Cost: trivial event-logging.

3. Recruit a cohort for ongoing eye-tracking. The v2-era eye-tracking sub-cohort gave us the most definitive evidence that the strip was being ignored. Eye-tracking is expensive but a recruited n = 15–25 cohort revisited quarterly is a research investment that pays back on every regulatory-UX change downstream.

4. Distinguish “engagement” from “completion.” The v1 audit trail conflated these two. Every user dismissed the modal and was logged as having engaged. A more disciplined engagement metric would have required a minimum read time and a minimum focus dwell, gating the Acknowledge button until those thresholds were met. The cost is friction; the gain is regulatory defensibility under a hypothetical ASIC review.

5. Plan for the next regulatory revision before shipping v1. The v3 absorbed two subsequent RG 268 revisions because the design was token-first. The v1 modal would have required a redesign for each revision. The implication for any regulatory-UX project: do not ship the first version of a pattern that you would not extend to the third version. If your v1 cannot absorb a revision, you are shipping technical debt regardless of how well the v1 performs.

6. Negative-result reporting as part of the project plan. Build the assumption into the project that some pre-launch projections will be wrong. Decide in advance how those will be reported. The AUM-didn’t-move result in Section 4 was reported because the team had agreed pre-launch that the business case would be assessed honestly post-launch regardless of direction. That agreement is what makes honest reporting possible; without it, the gravity is always toward selective narration of what worked.

How to read this case study

If you are evaluating this work for a hiring decision: this is what design leadership in a regulated context looks like over a long-enough window. Three versions, one workaround pattern I had to discover the hard way, one negative result I chose to report rather than hide, and one regulatory absorption pattern (the v3 design) that subsequently absorbed two RG 268 revisions without rebuild. The unit of analysis is not the v1 ship; it is the eighteen-month iteration arc.

The Microsoft brutal-review critique said: “I notice you have not shown me what this looked like six months after launch, when 10% of users had figured out the workaround that made your main flow irrelevant. Did you stay long enough to see that?” The answer is yes, and the workaround was the muscle-memory dismissal pattern that crashed the v1’s engagement quality at month three. The portfolio-elsewhere framing of the RG 268 work as a clean win without the v1 failure narrative is incomplete; this page is the complete version. The completeness is the senior-PD evidence.

Continue reading